About
Every feat comes with its receipt. Agentic Feats is a public record of what AI agents actually got done. Each entry is a feat: one real result, achieved by a real person with an agent like Claude Code, Codex, Cursor, Devin or OpenClaw, written in our own words, graded by how much proof exists, and linked to its source.
How a feat is made
No vendor marketing, no demos, no "imagine if". A feat is only listed when someone can point at what happened. The site is for the person asking "can an agent really do this for someone like me?", and for the journalist or researcher who needs sourced examples instead of claims.
A pipeline reads public sources every day: Hacker News, GitHub, YouTube, blogs and feeds, web search, and posts on X. It keeps the ones where someone did something and says what happened. A model drafts the card in our own words, never copying more than 25 words in a row. Every card is checked automatically before it is published, and the ones a check flags are reviewed by a person.
Evidence grades
Shown artifact linkedNumbers a specific number statedClaim claim only
The grade does not block a card. A claim is not bad; it has less proof. The methodology says how grades are given and how cards are chosen.
Neutral by rule
Agentic Feats has no relationship with any agent vendor. Sponsors never affect which cards appear, which are picks, or how they rank. A sponsored slot is always labeled and always outside the list of feats.
Workflow builders such as n8n, Make, and Zapier are out of scope. So are other directories: every card links to the original source.
Open data
Download every card as JSON or CSV from the open data page.